In this paper our arguments are based on the following hypotheses: A. In education, inevitably, scientific texts (professional, educational, popular etc) are the source to derive the scientific ideas/concepts for teaching. This constrain establishes, also inevitably, rather reproductive teaching-learning processes and practices. B. The ‘language’ of scientific texts, apart from its syntax and grammar difficulties, is convincing in a context dominated by logic, at least in its typical form. However, typical logical reasoning seems to leave students indifferent. Therefore, the limited success of constructivism is reasonable, as the logic seems to be the “key” of conceptual conflict. Furthermore, it is reasonable that for the education of children we need more familiar teaching-learning contexts, where the functions of scientific ideas are approached by narration or image, and anyway not simply by reason. C. The “narrative” approach is more familiar to the students but this does not change the fact that learning takes place through processes of reproduction. In any case, narrations using scientific ideas do almost not exist. We also think that a theory representing learning in a reproductive context does not exist. Both should be constructed. In this essay we present the main characteristics of a theory representing learning as a dialectic entity emerging in the context of social-cultural reproduction. We also attempt to support this theory analyzing qualitative some of our data collected from a long project of teaching Science in a theatrical / narrative manner.